Bookmark and Share

Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: dropped cases using mi impute mvn

From   [email protected] (Wesley D. Eddings, StataCorp)
To   [email protected]
Subject   Re: st: dropped cases using mi impute mvn
Date   Wed, 20 Oct 2010 18:37:01 -0500

Eric <[email protected]> asked about differences between the numbers of observations
in imputations created by -mi impute mvn-:

> The relevant lines of code are:
> mi set mlong
> mi impute mvn y1 y2 y3 y4 y5 y6  dummy1 dummy2 = x1 x2, add(3)
> This runs without error, but as you can see below, the number of cases is
> wrong for the three imputed data sets.  Somehow 730 cases are dropped.
> .	tab	_mi_m
> 		_mi_m       Freq.	Percent	Cum.
> 		0       2,812	31.04	       31.04
>		1       2,082	22.99	       54.03
>		2       2,082	22.99        77.01
>		3       2,082	22.99	      100.00
>	Total         9,058	       100.00

The -mi- commands are behaving as intended.  The -mi- commands offer several
storage styles, and Eric in his -mi set- command chose the -mlong- ("marginal
long") style, which stores only the 2082 incomplete observations in the
imputations.  The 730 complete observations are not repeated in the imputations
and are stored only in the original data.

Eric mentioned -ice-, which always repeats the complete observations in the
imputations.  The official -mi- style that repeats the complete observations is
-flong- ("full long").  All -mi- commands work regardless of style, but if Eric
would prefer to work in the -flong- style, he may convert to it with the -mi
convert- command.

For more information on -mi- styles, Eric may type

     . help mi glossary
     . help mi styles

-- Wes
[email protected]
*   For searches and help try:

© Copyright 1996–2018 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   Site index